L’interface est localisée ; les présentations et documents des Skills sont traduits automatiquement par IA dans votre langue. Le code, les commandes, chemins et termes techniques restent en version originale, et le document de l’auteur fait foi.
Explorer les Skills
Design et création · SKILL.md

recognition-pickers-and-palettes

recognition-pickers-and-palettes

Use this skill when designing pickers, dropdowns, comboboxes, autocompletes, command palettes, or any UI that lets the user select from a known set. Trigger when picking between a free-text field and a select; when designing a command palette; when reviewing why users keep typing wrong values into autocomplete inputs. Sub-aspect of `recognition-over-recall`…

Independent GitHub snapshot · Automated classification · Original documentation governs

Documentation de la Skill

GitHub

name: recognition-pickers-and-palettes description: 'Use this skill when designing pickers, dropdowns, comboboxes, autocompletes, command palettes, or any UI that lets the user select from a known set. Trigger when picking between a free-text field and a select; when designing a command palette; when reviewing why users keep typing wrong values into autocomplete inputs. Sub-aspect of recognition-over-recall; read that first.'

Pickers, palettes, and recognition-based selection

Selection UI is the most direct application of recognition-over-recall. Showing the user a list of valid options — visually scannable or filterable — beats requiring them to type from memory.

The picker ladder by set size

Different selection patterns suit different option counts:

Option countPatternNotes
2Toggle / switchVisual either-or
3–5Radio group / segmentedAll options visible
5–8Select dropdownVisible on click
8–20Grouped selectCategorical chunking helps
20+Combobox / typeaheadFilter-as-you-type
100+Search-first pickerSearch-driven; recents augment

Going lighter than the count permits forces unnecessary recall. Going heavier adds complexity without benefit.

Patterns

Combobox / typeahead

<combobox label="Country">
  <input placeholder="Search countries..." />
  <listbox>
    <option>United States</option>
    <option>United Kingdom</option>
    <!-- filtered as user types -->
  </listbox>
</combobox>

The user types partial recall; the system filters; the user recognizes the right option. Tolerates typos with fuzzy matching.

Command palette (cmd-K)

A palette that lists actions, navigation destinations, and search:

[ ⌘K opens palette ]

Search:
> Recents
  Create invoice
  Open settings
> Actions
  Search projects
  Invite teammate
> Navigation
  Go to Dashboard
  Go to Projects

Categorical groups + search + recents = recognition for nearly any command.

Recents at the top of pickers

<combobox label="Assignee">
  <listbox>
    <group label="Recent">
      <option>Maria Mendoza</option>
      <option>Lin Chen</option>
    </group>
    <group label="All members">
      <!-- alphabetical -->
    </group>
  </listbox>
</combobox>

Repeat-task speed via recents.

Suggestions and predictions

For inputs where the system can predict likely values:

<input
  name="email"
  type="email"
  autocomplete="email"
  list="email-suggestions"
/>
<datalist id="email-suggestions">
  <option value="user@example.com" />
  <option value="user@gmail.com" />
</datalist>

The browser surfaces matching options; the user recognizes and selects.

Visual pickers for inherently-visual content

For colors, dates, locations, etc., visual pickers are more recognition-friendly than coded inputs:

  • Color picker (visual swatch grid) beats hex code input.
  • Date picker (calendar) beats text date input.
  • Map picker (visual location) beats lat/long input.

Combobox accessibility

The WAI-ARIA Combobox pattern documents the keyboard interaction:

  • Tab enters the input.
  • Arrow keys navigate the listbox.
  • Enter selects the highlighted option.
  • Esc closes the listbox without selection.

Reach for an accessible combobox library (Radix, React Aria, Headless UI) rather than rolling your own — focus management and ARIA are easy to get wrong.

Anti-patterns

  • Free-text where a picker would do. Asking users to type "United States" exactly when a dropdown could show options.
  • Empty pickers that require typing to see anything (vs. showing options on focus).
  • Picker without filter for sets larger than ~10 items. Scrolling 200 items is slow.
  • No recents. Users repeatedly retype the same destinations.
  • Picker labels that don't match user vocabulary. Showing internal names instead of friendly names.

Heuristics

  1. The "what's available?" test. For each input, ask: how does the user know what valid options are? If they don't, you're requiring recall.
  2. The set-size match. Match the picker pattern to the option count.
  3. The recents-presence audit. For pickers used repeatedly, are recents surfaced? If not, every selection is a fresh recognition task.

Related sub-skills

  • recognition-over-recall (parent).
  • recognition-recents-and-suggestions — accelerating repeat tasks.
  • hicks-law-menus — picker design under decision-time constraints.
  • accessibility-operable — comboboxes are notorious accessibility-test surfaces.
Skills enregistrées